[lld] [ELF] Fix --lto-obj-path with the ThinLTO cache (PR #211062)

[Ben Dunbobbin via llvm-commits]

bd1976bris wrote:

> From looking at the comments in the code, it also looks like DTLTO could be affected too. I think the fix will work for DTLTO too but would be good to include this in the description. With the code structure as it is today I don't think we need a separate test for DTLTO. I've added some reviewers that have worked on DTLTO for any additional comments from that side.

Thanks for pinging me @smithp35. I agree that we don't need a DTLTO-specific test given the code structure.

This option isn't particularly relevant to the integrated distributed ThinLTO feature (DTLTO); however, I believe that it is used with the original distributed ThinLTO support, which uses separate thin-link, backend compilation, and link steps (as documented here: https://blog.llvm.org/2016/06/thinlto-scalable-and-incremental-lto.html).

I tried looking at the history of this option, but it isn't clear to me what the intended semantics are. This PR makes the output more predictable, which LGTM.

https://github.com/llvm/llvm-project/pull/211062